bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
1 2 3 4 Genome-Wide Association Study of Body Fat Distribution identifies Novel 5 Adiposity Loci and Sex-Specific Genetic Effects
6
7 Mathias Rask-Andersen*1, Torgny Karlsson1, Weronica E Ek1, Åsa Johansson*1
8
9
10 1Department of Immunology, Genetics and Pathology, Science for Life Laboratory,
11 Uppsala University. Box 256, 751 05, Uppsala, Sweden.
12 *Corresponding authors: [email protected] and
14
15
16
17
18
1 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
19 Body mass and body fat composition are of clinical interest due to their links to 20 cardiovascular- and metabolic diseases. Fat stored in the trunk has been 21 suggested as more pathogenic compared to fat stored in other compartments of 22 the body. In this study, we performed genome-wide association studies (GWAS) 23 for the proportion of body fat distributed to the arms, legs and trunk estimated 24 from segmental bio-electrical impedance analysis (sBIA) for 362,499 individuals 25 from the UK Biobank. A total of 97 loci, were identified to be associated with 26 body fat distribution, 40 of which have not previously been associated with an 27 anthropometric trait. A high degree of sex-heterogeneity was observed and 28 associations were primarily observed in females, particularly for distribution of 29 fat to the legs or trunk. Our findings also implicate that body fat distribution in 30 females involves mesenchyme derived tissues and cell types, female endocrine 31 tissues a well as several enzymatically active members of the ADAMTS family of 32 metalloproteinases, which are involved in extracellular matrix maintenance and 33 remodeling. 34
35 Overweight (body mass index [BMI] >25) and obesity (BMI>30) have reached
36 epidemic proportions globally1. Almost 40% of the world’s population are now
37 overweight2 and 10.8% are obese3. Obesity is set to become the world’s leading
38 preventable risk factor for disease and early death due to the increased risks of
39 developing type 2 diabetes, cardiovascular disease, and cancer4.
40
41 The distribution of adipose tissue to discrete compartments within the human body is
42 associated with differential risk for development of cardiovascular and metabolic
43 disease5. Body fat distribution of fat is also well known to differ between sexes. After
44 puberty, women accumulate fat in the trunk and limbs to a proportionally greater
45 extent compared to other parts of the body, while men accumulate a greater extent of
46 fat in the trunk6. Accumulation of adipose tissue around the viscera, the internal
47 organs of the body, has been shown to be associated with increased risk of disease in
2 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
48 both men and women7. In contrast, the preferential accumulation of adipose tissue in
49 the lower extremities, i.e. the hips and legs, has been suggested as a factor
50 contributing to the lower incidence of myocardial infarction and coronary death
51 observed in women during middle age8. The differential distribution of body fat
52 between sexes has been attributed to downstream effects of sex hormone secretion5.
53 However, the biological mechanisms that underlie body fat distribution have not been
54 fully elucidated.
55
56 BMI is commonly used as a proxy measurement of body adiposity in epidemiological
57 studies and in clinical practice. However, BMI is unable to discriminate between
58 adipose and lean mass, and between fat stored in different compartments of the body.
59 Other proxies that better represent distribution of body fat have also been utilized,
60 such as the waist-to-hip ratio (WHR), waist circumference (WC), and hip
61 circumference (HC). Through genome-wide association studies (GWAS), researchers
62 have identified hundreds of loci to be associated with proximal measurements of body
63 mass and body fat distribution such as BMI9, WHR10,11 and hip-, and waist
64 circumference11. Sex-stratified analyses have revealed sexual dimorphic effects at
65 twenty WHR-associated loci and 19 of these loci displayed stronger effects in
66 women12. Body fat mass has also been studied in GWAS by using bio-electrical
67 impedance analysis (BIA) and dual energy X-ray absorptiometry (DXA)13,14. BIA
68 measures the electrical impedance through the human body, which can be used to
69 calculate an estimate of the total amount of adipose tissue. The ‘gold standard’
70 method for measurements of body fat distribution is computed tomography (CT) or
71 magnetic resonance imaging (MRI). However, these methods are costly. A GWAS
72 has been performed for subcutaneous- and visceral adiposity, measured with
3 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
73 computed tomography scans, albeit in a relatively limited number of individuals
74 (N=10,577)15.
75
76 Developments in BIA technology has now allowed for cost-efficient segmental body
77 composition scans that estimate of the fat content of the trunk, arms and legs16 (Figure
78 1a). In this study, we used segmental BIA data on 362,499 participants of the UK
79 Biobank to study the genetic determinants of body fat distribution to the trunk, arms
80 and legs. For this purpose, we performed GWAS on the proportion of body fat
81 distributed to these compartments. We also performed sex-stratified analyses to
82 determine sex-specific effects and performed gene-sex interaction analyses to identify
83 effects that differ between men and women.
84
85 Results
86 The proportions of body fat distributed to the arms – arm fat ratio (AFR), the legs –
87 leg fat ratio (LFR) and the trunk – trunk fat ratio (TFR) were calculated by dividing
88 the fat mass per compartment with the total body fat mass for each participant (Figure
89 1a). We conducted a two-stage GWAS using data from the interim release of
90 genotype data in UK Biobank as a discovery cohort. Another set of participants, for
91 which genotype data were made available as part of the second release, was used for
92 replication. After removing non-Caucasians, genetic outliers and related individuals,
93 116,138 and 246,360 participants remained in the discovery and replication cohorts,
94 respectively. Basic characteristics of the discovery and replication cohorts are
95 presented in supplementary Table 1. Women were found to have higher total sBIA-
96 estimated fat mass compared to men in both the discovery and replication cohort, as
97 well as higher amount of fat in the arms and legs. Males had higher average
4 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
98 proportion of body fat located in the trunk compared to females (62.2% vs. 50.3%)
99 and women had a larger proportion of body fat located in the legs (39.7% vs. 28.1%).
100 While the total amount of adipose tissue in the arms was estimated to be higher in
101 women compared to men, the fraction of adipose tissue distributed to the arms were
102 similar. Several smaller differences between the discovery and replication cohorts
103 were present (supplementary Table 1), such as some slight differences in height and
104 age between men and women in the discovery and replication cohorts. These
105 differences most likely represent the 50,000 participants for the UK Biobank Lung
106 Exome Variant Evaluation (UK BiLEVE) project that were included in the first
107 release of genotyping data for ~150,000 participants, which were used as a discovery
108 cohort in this study. Selection for UK BiLEVE was conducted with specific
109 consideration to lung function which may reflect the differences in baseline
110 characteristics for this subset of the cohort. However, these differences are unlikely to
111 affect the results from our analyses.
112
113 Genome wide association study for body fat ratio
114 GWAS was performed for each of the three phenotypes (AFR, LFR and TFR) in the
115 whole discovery cohort (sex-combined) and when stratifying by sex (males and
116 females), adjusting for covariates as described in the method section. A total of
117 25,472,837 imputed SNPs, with MAF of at least 0.0001, were analyzed in the
118 discovery GWAS. LD score regression intercepts17 ranged from 1.00 to 1.03
119 (supplementary Figure 1, supplementary Table 2), and were used to adjust for
120 genomic inflation. We used the -clump function in PLINK18, in combination with
121 conditioning on the most significant SNP, to identify associations that were
122 independent within each GWAS as well as between GWAS for the three body fat
5 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
123 ratios (AFR, TFR, or LFR) or strata (males, females or sex-combined) (see methods).
124 In total, 133 independent associations at 114 loci were observed in the discovery
125 analyses (Figure 1c, Supplementary Figure 2-4, Supplementary Data 1). For each of
126 the independent associations, the leading SNP (the SNP that was most significant in
127 any of the phenotypes or strata) was taken forward for replication. In total, 97 of the
128 133 independent associations (Supplementary Data 1) replicated (P<0.05/133). Out of
129 these, 30 were associated with AFR, 42 with LFR and 65 with TFR in either the sex-
130 combined or sex-stratified analyses. There was substantial overlap in associated loci
131 between LFR and TFR loci (Figure 2) while associations with AFR overlapped to a
132 small degree. One locus in the vicinity of ADAMTSL3 was associated with all three
133 phenotypes.
134
135 In the sex-stratified GWAS, only six loci were associated (replication P-value <
136 0.05/133) with body fat ratios in males while 71 loci were observed in females. Sex-
137 heterogenous effects of associated variants were tested for using the GWAMA
138 software. We observed 42 independent body fat ratio-associated variants whose effect
139 differed between females and males (P < 5.15*10-4, Table 2). In all but two cases, the
140 effects were stronger, or only present in females. Stronger effects in males were
141 observed for two AFR-associated variants rs3812049 near SLC12A2, and rs11289753
142 near PLCE1 (Table 2).
143
144 LD score regression (LDSC) was used to estimate the fraction of variance of body fat
145 ratios that could be explained by SNPs, i.e. the SNP heritability17. SNP heritability
146 was higher in females compared to males for all traits and ranged from ~20% to
147 ~26% in females and from 12% to 16% in males (supplementary Table 2).
6 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
148
149 Phenotypic and genetic correlation between body fat ratios and anthropometric
150 phenotypes
151 Phenotypic correlations were assessed, in males and females separately, by
152 calculating squared semi-partial correlation coefficients from the ANOVA table of
153 two nested linear models that were adjusted for age and principal components, while
154 genetic correlations were estimated using cross-trait LD score regression19 (see
155 methods). Overall, the genetic and phenotypic correlations showed a large degree of
156 similarity (supplementary Table 3 and 4) and the correlation between the
157 anthropometric traits and the ratios were in the same direction for phenotypic and
158 genetic correlations for all correlations that reached the threshold for significance. In
159 females, BMI and WC was strongly correlated with AFR both with regards to
160 phenotypic (BMI - 78.9% and WC - 54.3%; squared semi-partial correlation
161 coefficients, see supplementary Table 3) and genetic (BMI - 79.2% and WC - 51,8%;
162 squared genetic correlation coefficients, see supplementary Table 4) correlations.
163 Height also contributed to a moderate degree in explaining the phenotypic variance in
164 LFR and TFR in females (16.0% and 25.3%) even though the genotypic correlation
165 between height and both LFR and TFR was even higher (42.3% and 60.8% variance
166 explained). In males, anthropometric traits contributed only to a small degree, up to
167 4.8%, in explaining the phenotypic variance in the ratios. This is also supported by the
168 lower genetic correlations between these traits with the highest correlation seen
169 between BMI and TFR (14.4% variance explained).
170
171 There was also a strong correlation between LFR and TFR in both males and females
172 (>82% of the variance explained in both males and females and for both genetic and
7 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
173 phenotypic correlations, supplementary Table 3 and 4). LFR and TFR were inversely
174 correlated, which agrees well with the large overlap in GWAS results for these
175 phenotypes and the fact that the effect estimates from the GWAS was in the opposite
176 direction for LFR and TFR (Supplementary Data 1). In contrast, AFR appeared to be
177 more independent as only a low amount of phenotypic and genetic correlation was
178 observed between AFR and the other two traits (supplementary Table 3 and 4).
179
180 Overlap with findings from previous GWAS
181 Body fat ratio-associated SNPs were tested for overlap with associations from
182 previous GWAS for anthropometric traits by determining LD with entries from
183 GWAS-catalog20. We used a strict cut-off of R2 < 0.1 to distinguish novel findings. In
184 total, we identify 40 body fat ratio-associated signals that have not previously been
185 associated with an anthropometric trait (Table 1). Forty body fat ratio associated
186 signals overlapped with previously identified height-associated loci21,22
187 (Supplementary Data 1) and the majority of these signals were associated with TFR
188 and/or LFR (36 out of 40). For AFR, the strongest associations were observed at well-
189 known BMI and adiposity-associated loci such as: FTO, MC4R, TMEM18, SEC16B
190 and TFAP2B (Supplementary Data 1), which agrees with the strong correlations
191 (genotypic and phenotypic) between AFR and BMI.
192
193 We also compared the direction of the effects for overlapping GWAS results by
194 estimating the effects of leading body fat ratio-associated SNPs on the respective
195 overlapping anthropometric traits in the discovery cohort. The effects of TFR-
196 associated SNPs were directionally consistent with effects on height and
197 WHRadjBMI, while the effects were the opposite for LFR. The direction of effects
8 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
198 for AFR-associated SNPs were consistent with effects on BMI-, WC- and WHR
199 (supplementary Table 5), which agrees with the strong genetic and phenotypic
200 correlation between AFR and these phenotypes.
201
202 Among the novel loci, five overlapped with traits associated with cardiovascular
203 disease: two loci near ANKDD1B and KAT8 that have been associated with LDL
204 cholesterol23 and triglycerides24, respectively; and three loci near KCNH2, XKR6 and
205 BMP2 that have been associated with QT interval25–27, thickness of the carotid intima
206 media28 and QRS complex29, respectively (supplementary Table 6).
207
208
209 Functional annotation of associated loci
210 Functional annotation of the GWAS loci was performed by identifying overlap with
211 eQTLs from the Genotype-Tissue Expression (GTEx) project30 and by identifying
212 potentially deleterious missense variants in LD (R2 > 0.8) with our leading SNPs (see
213 method section). In total, 31 body fat ratio-associated loci overlapped with an eQTL
214 (Supplementary Data 2), and 11 leading SNPs were in LD with a potentially
215 deleterious missense variant (Table 3). The probability for functional effects of
216 missense variants has been predicted by sequence analyses31,32 and Polyphen and
217 SIFT-scores were used to assess the deleteriousness of the variants. Plausible causal
218 missense variants were found in ACAN, ADAMTS17, FGFR4 and ADAMTS10, where
219 the leading SNPs were predicted to have functional effects (Table 3). Rs315855,
220 within FGFR4, has also previously been shown to be associated with progression of
221 cancer33,34 and to affect insulin secretion in vitro35.
222
9 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
223 Enrichment analyses
224 To identify the functional roles and tissue specificity of associated variants, we
225 performed enrichment analyses with DEPICT (Data-driven Expression Prioritized
226 Integration for Complex Traits36, see method section). In these analyses we used
227 summary statistics from sex-stratified GWAS on the combined cohort (195,043
228 women and 167,408 men) in order to maximize statistical power. Enrichment was
229 only detected for TFR- and LFR-associated genes in females as well as LFR-
230 associated genes in males (supplementary Data 3). We identified 212 enriched gene
231 sets and there was a substantial overlap of enriched gene sets between TFR- and LFR-
232 in females as well as moderate overlap with LFR-associated gene sets in males
233 (supplementary Figure 5). The large fraction of overlapping gene sets between LFR
234 and TFR in females agrees well with the large overlap in GWAS signals. Gene sets
235 related to bone morphology and skeletal development (abnormal skeleton
236 morphology, short limbs, decreased length of long bones, skeletal system
237 morphogenesis) were among the most strongly associated with both LFR and TFR.
238 We also find the 'TGFβ signaling pathway' gene set to be enriched for genes within
239 the TFR and LFR-associated loci in females, as well as TGFβ downstream mediators:
240 'SMAD1-', 'SMAD2-', 'SMAD3-' and 'SMAD7 protein-protein interaction
241 subnetworks' (supplementary Data 3).
242
243 The results from enrichment analyses were compared with results from previous
244 GWAS for height21, BMI9 and WHRadjBMI12. Gene-sets that were enriched for LFR
245 and TFR-associated genes were compared to results from previous GWAS for
246 height21, BMI9 and WHRadjBMI12. Substantial overlap of enriched gene sets was
247 observed between LFR/TFR- with height and WHRadjBMI (Figure 3a). In contrast,
10 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
248 BMI associated biological processes overlapped only to a marginal extent: 2% of all
249 gene sets that enriched for BMI-associated genes were also enriched for LFR- and
250 TFR-associated genes.
251
252 Tissue enrichment was observed for LFR and TFR-associated genes in females
253 (Figure 3b-c) in gene sets related to female reproduction, musculoskeletal systems,
254 chondrocytes, mesenchymal stem cells, and fibroblasts. For TFR, DEPICT also
255 revealed enrichment of genes associated with adipose tissue cells as well as endocrine
256 and cardiovascular systems (Figure 3). Tissue enrichment was not seen for the other
257 traits or strata.
258
259 Discussion
260 In this study, we performed GWAS on distribution of body fat to different
261 compartments of the human body and identified and replicated 97 independent
262 associations of which 40 have not been associated with any adiposity related
263 phenotype previously. In contrast to previous studies, we have not addressed the total
264 amount of fat but rather the fraction of the total body fat mass that is located in the
265 arms, legs and trunk. Body fat distribution is well-known to differ between males and
266 females, which we also clearly show in our study. We also show that the genetic
267 effects that influence fat distribution are stronger in females compared to males.
268 These results are consistent with previous GWAS that have revealed sexual
269 dimorphisms in genetic loci for adiposity-related phenotypes, such as waist-
270 circumference and waist-to-hip ratio10,37,38. Phenotypic and genetic correlations, as
271 well as results from GWAS and subsequent enrichment analyses, also revealed that
272 the amount of fat stored in the arms in females is highly correlated with BMI and WC.
11 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
273 This suggests that the proportion of fat stored in the arms will generally increase with
274 increased accumulation of body mass and adipose tissue. In contrast, males exhibited
275 moderate- to weak phenotypic and genetic correlations between the distribution of fat
276 to different parts of the body and anthropometric traits, which indicates that the
277 proportions of body fat mass in different compartments of the male body remains
278 more stable as body mass and body adiposity increases. Among the three phenotypes
279 analyzed in this study LFR and TFR were inversely correlated in both males and
280 females. This suggests that LFR and TFR to a large extent describe one trait, i.e. the
281 distribution of adipose tissue between these two compartments which is further
282 supported by the large overlap in GWAS loci between the two phenotypes. In
283 contrast, AFR was only weakly correlated with the other two traits.
284
285 Tissue enrichment revealed an important role in body fat distribution in females for
286 mesenchyme derived tissues: i.e. adipose and musculoskeletal tissues; as well as
287 tissues related to female reproduction. This suggests that the distribution of fat to the
288 legs and trunk in females is mainly driven by the effect of female gonadal hormones
289 on mesenchymal progenitors of musculoskeletal and adipose tissues. Enrichment
290 analyses also showed that LFR and TFR have unique features that separates them
291 from other anthropometric measurements, which was indicated by the portion of
292 LFR/TFR-associated gene sets that did not overlap with height-, BMI or
293 WHRadjBMI-associated gene sets. However, there was also an overlap in the
294 functional aspects between these traits with both height and WHRadjBMI. This is
295 indicated by the tissue enrichment profile for LFR/TFR-associated genes, which
296 shares features with tissue enrichments reported for height in previous GWAS39:
297 height-associated genes were strongly enriched in musculoskeletal tissue types with
12 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
298 additional enrichment in cardiovascular and endocrine tissue types; as well as
299 WHRadjBMI12: associated genes were enriched in adipocytes and adipose tissue
300 subtypes. Of particular note, we did not identify any enrichment of body fat ratio-
301 associated genes in CNS tissue gene sets in contrasts to enrichment analyses in
302 previous GWAS for BMI where CNS have been indicated to play a prominent role in
303 obesity susceptibility9.
304
305 In the GWAS for LFR and TFR in females, we find that several genes that highlight
306 the influence of biological processes related to the interaction between cells and the
307 extracellular matrix (ECM), as well as ECM-maintenance and remodeling. These
308 include ADAMTS2, ADAMTS3, ADAMTS10, ADAMTS14, and ADAMTS17, which
309 encode extracellular proteases that are involved in enzymatic remodeling of the ECM.
310 Two leading SNPs were in LD with potentially damaging missense mutations in
311 ADAMTS10 and ADAMTS17 and two other GWAS signals overlapped with eQTLs
312 for ADAMTS14 and ADAMTS3. In addition, possibly deleterious missense mutations
313 in LD with our leading GWAS SNPs were also found for VCAN and ACAN. Both
314 VCAN and ACAN encode chondroitin sulfate proteoglycan core proteins that
315 constitute structural components of the extracellular matrix, particularly in soft
316 tissues40. These proteins also serve as major substrates for ADAMTS proteinases41.
317 ECM forms the three-dimensional support structure for connective and soft tissue. In
318 fat tissue, the ECM regulates adipocyte expansion and proliferation42. Remodeling of
319 the ECM is required to allow for adipose tissue growth and this is achieved through
320 enzymatic processing of extracellular molecules such as proteoglycans, collagen and
321 hyaluronic acid. For example, ADAMTS2, 3 and 14 act as procollagen N-
322 propeptidases that mediate the maturation of triple helical collagen fibrils43,44. We
13 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
323 therefore propose that effects of genetic variation in biological systems involved in
324 ECM-remodeling as a factor underlying normal variation in female body fat
325 distribution.
326
327 Conclusions
328 In summary, GWAS of body fat distribution determined by sBIA reveals a genetic
329 architecture that influences distribution of adipose tissue to the arms, legs and trunk.
330 Genetic associations and effects clearly differ between sexes, in particular for
331 distribution of adipose tissue to the legs and trunk. The distribution of body fat in
332 women has been previously been suggested as a causal factor leading to lower risk of
333 cardiovascular and metabolic disease, as well as cardiovascular mortality for women
334 in middle age5 and genetic studies have identified SNPs that are associated with a
335 ‘favorable’ body fat distribution45, i.e. with higher BMI but lower risk of
336 cardiovascular and metabolic disease. The capacity for peripheral adipose storage has
337 been highlighted as one of the components underlying this phenomenon45. Resolving
338 the genetic determinants and mechanisms that lead to a favorable distribution of body
339 fat may help in risk assessment and in identifying novel venues for intervention to
340 prevent or treat obesity-related disease.
341
342 Author contributions
343 MRA, TK, WE and ÅJ conceived of and designed the study. Analysis was performed
344 by MRA and WE under supervision by ÅJ. MRA analyzed the data and wrote the first
345 draft of the manuscript. All authors contributed to the final version of the manuscript.
346
347 Acknowledgements
14 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
348 We are grateful to the participants and staff of the UK Biobank. Access to UK
349 Biobank genetic and phenotypic data was granted under application no. 15152.
350 Computations were performed on the computational cluster at the Uppsala
351 Multidisciplinary Center for Advanced Computational Science (UPPMAX) under
352 projects b2016021 and b2017066. The work was supported by grants from the
353 Swedish Society for Medical Research (SSMF), the Kjell and Märta
354 Beijers Foundation, Göran Gustafssons Foundation, the Swedish Medical Research
355 Council (Project Number 2015-03327), the Marcus Borgström Foundation, and the
356 Åke Wiberg Foundation.
357
358 online Methods
359 UK Biobank participants
360 The first release of imputed genotype data from UK Biobank (N = 152,249) was used
361 as a discovery cohort, and imputed genotype data from an unrelated set of participants
362 from the third genotype release (N = 326,565) as a replication cohort. Participants
363 who self-reported as being of British descent (data field 21000) and were classified as
364 Caucasian by principal component analysis (data field 22006) were included in the
365 analysis. Genetic relatedness pairing was provided by the UK Biobank (Data field
366 22011). Participants were removed due to relatedness based on kinship data
367 (estimated genetic relationship > 0.044), poor genotyping call rate (<95%), high
368 heterozygosity (Data field 22010), or sex-errors (Data filed 22001). After filtering,
369 116,138 participants remained in the discovery cohort and 246,361 in the replication
370 cohort and were included in downstream analyses.
371
372 Ethics
15 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
373 Ethical approval to collect participant data was given by the North West Multicentre
374 Research Ethics Committee, the National Information Governance Board for Health
375 & Social Care, and the Community Health Index Advisory Group. UK Biobank
376 possesses a generic Research Tissue Bank approval granted by the National Research
377 Ethics Service (http://www.hra.nhs.uk/), which lets applicants conduct research on
378 UK Biobank data without obtaining separate ethical approvals. Access to UK
379 Biobank genetic and phenotypic data was granted under application no. 15152. All
380 participants provided signed consent to participate in UK Biobank 46.
381
382 Genotyping, imputations and QC.
383 Genotyping in the discovery cohort had been performed on two custom-designed
384 microarrays: referred to as UK BiLEVE and Axiom arrays, which genotyped 807,411
385 and 820,967 SNPs, respectively. Imputation had been performed using UK10K47 and
386 1000 genomes phase 348 as reference panels. Prior to analysis, we filtered SNPs based
387 on call rate (--geno 0.05), HWE (P-value > 10-20, MAF (--maf 0.0001) and imputation
388 quality (Info > 0.3) resulting in 25,472,837 SNPs in the discovery cohort. The third
389 release of data from the UK Biobank contained genotyped and imputed data for
390 488,366 participants (partly overlapping with the first release). For our replication
391 analyses, we included an independent subset that did not overlap with the discovery
392 cohort. Genotyping in this subset was performed exclusively on the UK Biobank
393 Axiom Array. This dataset included 47,512,111 SNPs that were filtered based on
394 HWE (P<10-20), call rate (--geno 0.05), (Info >0.3) and MAF (--maf 0.0001). All
395 genomic positions are in reference to hg19/build 37.
396
397 Phenotypic measurements
16 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
398 The phenotypes used in this study derive from impedance measurements produced by
399 the Tanita BC418MA body composition analyzer. Participants were barefoot, wearing
400 light indoor clothing, and measurements were taken with participants in the standing
401 position. Height and weight were entered manually into the analyzer before
402 measurement. The Tanita BC418MA uses eight electrodes: two for each foot and two
403 for each hand. This allows for five impedance measurements: whole body, right leg,
404 left leg, right arm and left arm (Figure 1a.). Body fat for the whole body and
405 individual body parts had been calculated using a regression formula, that was derived
406 from reference measurements of body composition by DXA (Figure 1b) in Japanese
407 and Western subjects. This formula uses weight, age, height and impedance
408 measurements49 as input data. Arm, and leg fat masses were averaged over both
409 limbs. Arm, leg, and trunk fat masses were then divided by the total body fat mass to
410 obtain the ratios of fat mass for the arms, legs and trunk, i.e. what proportion of the
411 total fat in the body is distributed to each of these compartments. These variables
412 were analyzed in this study and were named: arm fat ratio (AFR), leg fat ratio (LFR),
413 and trunk fat ratio (TFR).
414
415 Assessing the relationship between adipose tissue ratios and anthropometric traits
416 Phenotypic correlations between fat distribution ratios and anthropometric traits were
417 estimated by calculating semi-partial correlation coefficients for males and females
418 separately , using anova.glm in R. Adipose tissue ratios (AFR, LFR or TFR) were set
419 as the response variable. BMI, waist circumference, waist circumference adjusted for
420 BMI, waist-to-hip ratio, height, or one of the other ratios were included as the last
421 term in a linear model that included, age and principal components as covariates. The
422 reduction in residual deviance, i.e., the reductions in the residual sum of squares as
17 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
423 BMI, waist circumference, waist circumference adjusted for BMI, waist-to-hip ratio,
424 height, or one of the other ratios was added to the model, is presented as percentages
425 of the total deviance of the null model in supplementary Table 3.
426
427 Genome wide association study for body fat ratio
428 A two-stage GWAS was performed using a discovery and a replication cohort. Body
429 fat ratios were adjusted for age, age squared and normalized by rank-transformation
430 separately in males and females using the rntransform function included in the
431 GenABEL library50. GWAS was performed in PLINK v1.90b3n18 using linear
432 regression models with AFR, LFR, and TFR as the response variables and the SNPs
433 as predictor variables. A batch variable was used as covariate in the GWAS for the
434 discovery analyses to adjust for genotyping array (Axiom and BiLEVE). We also
435 included age, the first 15 principal components and sex (in the sex-combined
436 analyses) as covariates in the GWAS. LD score regression intercepts (see further
437 information below), calculated using ldsc17, were used to adjust for genomic inflation,
438 by dividing the square of the t-statistic for each tested SNP with the LD-score
439 regression intercept for that GWAS, and then calculating new p-values based on the
440 adjusted t-statistic. We used a threshold of P<10-7, after adjusting for LD score
441 intercept, as threshold for significance in the discovery cohort.
442
443 The -clump function in PLINK was used to identify the number of independent
444 signals in each GWAS. This function groups associated SNPs based on the linkage
445 disequilibrium (LD) pattern. The parameters for clumps were set to: -clump-p1 1*10-
446 7, clump-p2 1*10-7, clump-r2 0.10 and --clump-kb 1000. This function groups SNPs
447 within one million base pairs that were associated with the trait at p < 1*10-7. Several
18 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
448 associations were found in more than one of the three body fat ratios (AFR, TFR, or
449 LFR) or strata (males, females or sex-combined) and different leading SNPs were
450 observed for different traits and strata at several loci. To assess whether these
451 represented the same signal, we assessed the LD between overlapping leading SNPs
452 in PLINK. SNPs in low LD (R2-value < 0.05) were considered to represent
453 independent signals. For leading SNPs where we could not exclude that they were in
454 LD (R2-value >= 0.05) we performed conditional analysis in PLINK, conditioning on
455 the most significant SNP across all phenotypes and strata. Associations with a P<
456 1*10-7 after conditioning for the most significant SNP were considered as being
457 independent signals. For each independent signal, the leading SNP (lowest p-value)
458 was taken forward for replication. Meta analyses of results from the discovery and
459 replication cohorts was performed with the METAL software51 for all independent
460 associations that were taken forward for replication.
461
462 SNP heritability, and genetic correlations
463 We estimated SNP heritability and genetic correlations using LD score regression
464 (LDSC), implemented in the 'ldsc' software package17. LDSC uses LD patterns and
465 summary stats from GWAS as input. For genetic correlations, we performed
466 additional sex-stratified GWAS in the UK biobank (using the same covariates as for
467 the ratios) for standard anthropometric traits, BMI, height, WC, WHR, WCadjBMI
468 and WHRadjBMI, in the discovery cohort. GWAS summary stats were filtered for
469 SNPs included in HapMap3 to reduce likelihood of bias induced by poor imputation
470 quality. After this filtering, 1,164,192 SNPs remained for LDSC analyses. LD scores
471 from the European data of the 1000 Genomes project (including LD patterns for all
472 the HapMap3 SNPs) for use with LDSC were downloaded from the Broad institute at:
19 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
473 https://data.broadinstitute.org/alkesgroup/LDSCORE/eur_w_ld_chr.tar.bz2. Genetic
474 correlations between the three body fat ratios and anthropometric traits were assessed
475 by cross-trait LD score regression.
476
477 Overlap with findings from previous GWAS
478 Leading SNPs from all independent signals in our analyses were cross referenced
479 with the NHGRI-EBI catalog of published genome-wide association studies (GWAS
480 Catalog – data downloaded on 23 April 2018) 20 to determine if body fat ratio-
481 associated signals overlapped with previously identified anthropometric associations
482 from previous GWAS. We used a cutoff of R2 < 0.1 between SNPs from our analyses
483 and anthropometric trait-associated SNPs (P< 5*10-8) from GWAS catalog to
484 determine if an association was novel. LD between data in the GWAS catalogue and
485 our leading SNPs were calculated using PLINK v1.90b3n18. In addition, leading SNPs
486 at Body fat ratio-associated loci that potentially overlapped (R2 > 0.1) with signals
487 from previous GWAS were tested for association with standard anthropometric traits
488 (BMI, height, WC, WHR, WCadjBMI and WHRadjBMI) in the UK biobank
489 discovery cohort using PLINK v1.90b3n18 through linear regression modelling and
490 including sex, age a batch variable and 15 principal components as covariates. Here, a
491 P-value of P < 1e-7 was considered significant.
492
493 Functional annotation of associated loci
494 Associated loci were investigated for overlap with eQTLs from the GTEx project 30.
495 The threshold for significance for the eQTLs was set to 2.3*10-9 in agreement with
496 previous studies52. The strongest associated SNP for each tissue and gene in the GTEx
497 dataset was identified. We then estimated the LD between the top eQTL SNPs and the
20 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
498 leading SNPs for each independent association from our analysis. If a SNPs from our
499 analyses were in LD (R2 > 0.8) with a leading eQTL SNP the two signals were
500 considered overlapping.
501
502 Leading SNPs, and all SNPs in LD (R2 > 0.8) with a leading SNPs from our analyses
503 (LD determined in the UK biobank cohort in PLINK) were cross-referenced with
504 dbSNP (human 9606 b150) in order to identify potentially deleterious intragenic
505 variants in LD (R2 > 0.8) with body fat ratio-associated variants. Polyphen and SIFT-
506 scores for the missense variants (extracted from Ensembl - www.ensembl.org) were
507 used to assess the deleteriousness of the variants.
508
509 Enrichment analysis
510 To identify the functional roles and tissue specificity of associated variants, we
511 performed tissue and gene-set enrichment analyses using DEPICT36. For the gene-set
512 enrichment in DEPICT, gene expression data from 77,840 samples have been used to
513 predict gene function for all genes in the genome based on similarities in gene
514 expression. In comparison to standard enrichment tools that apply a binary definition
515 to define membership in a set of genes that have been associated with a biological
516 pathway or functional category (genes are either included or not included), in
517 DEPICT, the probability of a gene being a member of a gene set has instead been
518 estimated based on correlation in gene expression. This membership probability to
519 each gene set has been estimated for all genes in the human genome and the
520 membership probabilities for each gene have been designated 'reconstituted' gene sets.
521 A total of 14,461 reconstituted gene sets have been generated which represent a wide
522 set of biological annotations (Gene Ontology [GO], KEGG, REACTOME,
21 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
523 Mammalian Phenotype [MP], etc.). For tissue enrichment in DEPCIT, microarray
524 data from 37,427 human tissues have been used to identify genes with high
525 expression in different cells and tissues. When performing the tissue enrichment
526 analyses in DEPICT, the most significant GWAS hits are scanned for enrichment of
527 genes that are highly expressed in any cells or tissues.
528
529 For the enrichment analyses, we performed sex-stratified GWAS for AFR, LFR and
530 TFR on a combined set of the UK Biobank cohort, in order to achieve high power for
531 enrichment analyses, including 195,043 females and 167,408 males. The 'clump'
532 functionality in PLINK is used to determine associated loci. The p-value cut off for
533 association for 'clump' was set at P < 10-7. In the enrichment analyses, DEPICT
534 assesses whether the reconstituted gene sets are enriched for genes within trait-
535 associated loci36. The False Discovery Rate (FDR)53 was used to adjust for multiple
536 testing. Twelve analyses were run in total (tissue enrichment and gene-set
537 enrichment) and FDR<0.05/12 was considered significant.
538
539 Interaction between SNPs and sex
540 We used the GWAMA software54 to test for heterogenous effects of associated SNPs
541 between sexes. In GWAMA, fixed-effect estimates of sex-specific and sex-combined
542 beta coefficients and standard errors are calculated from GWAS summary statistics to
543 test for heterogeneous allelic effects between females and males. GWAMA obtains a
544 test-statistic by subtracting the sex-combined squared t-statistic from the sum of the
545 two sex-specific squared t-statistics. This test statistic is asymptotically chi-square-
546 distributed and equivalent to a normal z-test of the difference in allelic effects
547 between sexes. SNPs that were significant in the replication analyses in any of the
22 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
548 strata were tested for heterogeneity between sexes in the replication cohort.
549 Bonferroni correction was used to correct for multiple testing and p-values < 0.05/97
550 were considered to be significant.
551
552
553 Data availability
554 The data that support the findings of this study are available from UK Biobank
555 (http://www.ukbiobank.ac.uk/about-biobank-uk/). Restrictions apply to the
556 availability of these data, which were used under license for the current study (project
557 no. 15152). Data is available for bona fide researchers upon application to the UK
558 Biobank. Summary statistics from all association tests are available for download at:
559 https://myfiles.uu.se/ssf/s/readFile/share/3993/1270878243748486898/publicLink/G
560 WAS_summary_stats_ratios.zip.
561
562 References
563 1. Afshin, A. et al. Health Effects of Overweight and Obesity in 195 Countries
564 over 25 Years The. N. Engl. J. Med. 377, 13–27 (2017).
565 2. Ng, M. et al. Global, regional, and national prevalence of overweight and
566 obesity in children and adults during 1980-2013: A systematic analysis for the
567 Global Burden of Disease Study 2013. Lancet 384, 766–781 (2014).
568 3. Di Cesare, M. et al. Trends in adult body-mass index in 200 countries from
569 1975 to 2014: A pooled analysis of 1698 population-based measurement
570 studies with 19.2 million participants. Lancet 387, 1377–1396 (2016).
571 4. Pi-Sunyer, F. X. et al. Clinical guidelines on the identification, evaluation, and
572 treatment of overweight and obesity in adults: The Evidence Report. (1998).
23 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
573 5. Link, J. C. & Reue, K. The Genetic Basis for Sex Differences in Obesity and
574 Lipid Metabolism. Annu. Rev. Nutr. 37, annurev-nutr-071816-064827 (2017).
575 6. Bouchard, C., Despreis, J. P. & Maurieige, P. Genetic and nongenetic
576 determinants of regional fat distribution. Endocr. Rev. 14, 72–93 (1993).
577 7. Wajchenburg, B. L. Subcutaneous and Visceral Adipose Tissue : Their
578 Relation to the Metabolic Syndrome. Endocr. Rev. 21, 697–738 (2014).
579 8. Tunstall-Pedoe, H. Myth and paradox of coronary risk and the menopause.
580 Lancet 351, 1425–1427 (1998).
581 9. Locke, A. E. et al. Genetic studies of body mass index yield new insights for
582 obesity biology. Nature 518, 197–206 (2015).
583 10. Heid, I. M. et al. Meta-analysis identifies 13 new loci associated with waist-hip
584 ratio and reveals sexual dimorphism in the genetic basis of fat distribution. Nat.
585 Genet. 42, 949–960 (2010).
586 11. Wen, W. et al. Genome-wide association studies in East Asians identify new
587 loci for waist-hip ratio and waist circumference. Sci. Rep. 6, 17958 (2016).
588 12. Shungin, D. et al. New genetic loci link adipose and insulin biology to body fat
589 distribution. Nature 518, 187–196 (2015).
590 13. Kilpeläinen, T. O. et al. Genetic variation near IRS1 associates with reduced
591 adiposity and an impaired metabolic profile. Nat. Genet. 43, 753–760 (2011).
592 14. Lu, Y. et al. New loci for body fat percentage reveal link between adiposity
593 and cardiometabolic disease risk. Nat. Commun. 7, (2016).
594 15. Fox, C. S. et al. Genome-wide association for abdominal subcutaneous and
595 visceral adipose reveals a novel locus for visceral fat in women. PLoS Genet. 8,
596 (2012).
597 16. Mally, K., Trentmann, J., Heller, M. & Dittmar, M. Reliability and accuracy of
24 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
598 segmental bioelectrical impedance analysis for assessing muscle and fat mass
599 in older Europeans: A comparison with dual-energy X-ray absorptiometry. Eur.
600 J. Appl. Physiol. 111, 1879–1887 (2011).
601 17. Bulik-Sullivan, B. et al. LD score regression distinguishes confounding from
602 polygenicity in genome-wide association studies. Nat. Genet. 47, 291–295
603 (2015).
604 18. Weeks, J. P. plink: An R Package for Linking Mixed-Format Tests Using IRT-
605 Based Methods. J. Stat. Softw. 35, 1–33 (2010).
606 19. Bulik-Sullivan, B. et al. An atlas of genetic correlations across human diseases
607 and traits. Nat. Genet. 47, 1236–1241 (2015).
608 20. MacArthur, J. et al. The new NHGRI-EBI Catalog of published genome-wide
609 association studies (GWAS Catalog). Nucleic Acids Res. 45, D896–D901
610 (2017).
611 21. Wood, A. R. et al. Defining the role of common variation in the genomic and
612 biological architecture of adult human height. Nat. Genet. 46, 1173–86 (2014).
613 22. He, M. et al. Meta-analysis of genome-wide association studies of adult height
614 in East Asians identi fi es 17 novel loci. Hum. Mol. Genet. 24, 1791–1800
615 (2017).
616 23. Kathiresan, S. et al. Six new loci associated with blood low-density lipoprotein
617 cholesterol, high-density lipoprotein cholesterol or triglycerides in humans.
618 Nat. Genet. 40, 189–197 (2008).
619 24. Teslovich, T. M. et al. Biological, clinical and population relevance of 95 loci
620 for blood lipids. Nature 466, 707–713 (2010).
621 25. Newton-Cheh, C. et al. Common variants at ten loci influence QT interval
622 duration in the QTGEN Study. Nat. Genet. 41, 399–406 (2009).
25 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
623 26. Holm, H. et al. Several common variants modulate heart rate, PR interval and
624 QRS duration. Nat. Genet. 42, 117–122 (2010).
625 27. Arking, D. E. et al. Genetic association study of QT interval highlights role for
626 calcium signaling pathways in myocardial repolarization. Nat. Genet. 46, 826–
627 836 (2014).
628 28. Bis, J. C. et al. Meta-analysis of genome-wide association studies from the
629 CHARGE consortium identifies common variants associated with carotid
630 intima media thickness and plaque. Nat. Genet. 43, 940–948 (2011).
631 29. van der Harst, P. et al. 52 Genetic Loci Influencing Myocardial Mass. J. Am.
632 Coll. Cardiol. 68, 1435–1448 (2016).
633 30. Aguet, F. et al. Genetic effects on gene expression across human tissues.
634 Nature 550, 204–213 (2017).
635 31. Ng, P. C. & Henikoff, S. SIFT : predicting amino acid changes that affect
636 protein function. 31, 3812–3814 (2003).
637 32. Adzhubei, I., Jordan, D. M. & Sunyaev, S. R. Predicting functional effect of
638 human missense mutations using PolyPhen-2. Current Protocols in Human
639 Genetics (2013). doi:10.1002/0471142905.hg0720s76
640 33. Allele, A. et al. Cancer Progression and Tumor Cell Motility Are Associated
641 with the FGFR4. 840–847 (2002).
642 34. Frullanti, E. et al. Meta and pooled analyses of FGFR4 Gly388Arg
643 polymorphism as a cancer prognostic factor. Eur. J. Cancer Prev. 20, 340–347
644 (2011).
645 35. Ezzat, S. et al. The Cancer-associated FGFR-G388R Polymorphism Enhances
646 Pancreatic Insulin Secretion and Modifies the Risk of Diabetes. Cell Metab.
647 929–940 (2013). doi:10.1016/j.cmet.2013.05.002
26 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
648 36. Pers, T. H. et al. Biological interpretation of genome-wide association studies
649 using predicted gene functions. Nat. Commun. 6, (2015).
650 37. Randall, J. C. et al. Sex-stratified Genome-wide Association Studies Including
651 270,000 Individuals Show Sexual Dimorphism in Genetic Loci for
652 Anthropometric Traits. PLoS Genet. 9, (2013).
653 38. Winkler, T. W. et al. The Influence of Age and Sex on Genetic Associations
654 with Adult Body Size and Shape: A Large-Scale Genome-Wide Interaction
655 Study. PLoS genetics 11, (2015).
656 39. Wood, A. R. et al. Defining the role of common variation in the genomic and
657 biological architecture of adult human height. Nat. Genet. 46, (2014).
658 40. Wight, T. N. Versican: A versatile extracellular matrix proteoglycan in cell
659 biology. Curr. Opin. Cell Biol. 14, 617–623 (2002).
660 41. Binder, M. J., McCoombe, S., Williams, E. D., McCulloch, D. R. & Ward, A.
661 C. The extracellular matrix in cancer progression: Role of hyalectan
662 proteoglycans and ADAMTS enzymes. Cancer Lett. 385, 55–64 (2017).
663 42. Lin, D., Chun, T. & Kang, L. Adipose extracellular matrix remodelling in
664 obesity and insulin. Biochem. Pharmacol. 119, 8–16 (2016).
665 43. Bekhouche, M. et al. Determination of the substrate repertoire of ADAMTS2,
666 3, and 14 significantly broadens their functions and identifies extracellular
667 matrix organization and TGF-β signaling as primary targets. FASEB J. 30,
668 1741–1756 (2016).
669 44. Kelwick, R., Desanlis, I., Wheeler, G. N. & Edwards, D. R. The ADAMTS ( A
670 Disintegrin and Metalloproteinase with Thrombospondin motifs ) family.
671 (2015). doi:10.1186/s13059-015-0676-3
672 45. Yaghootkar, H. et al. Genetic Evidence for a Link Between Favorable
27 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
673 Adiposity and Lower Risk of Type 2 Diabetes , Hypertension , and Heart
674 Disease. 65, 2448–2460 (2016).
675 46. UK Biobank. UK Biobank ethics and governance framework version 3.0. 0, 3–
676 18 (2007).
677 47. Walter, K. et al. The UK10K project identifies rare variants in health and
678 disease. Nature 526, 82–89 (2015).
679 48. Auton, A. et al. A global reference for human genetic variation. Nature 526,
680 68–74 (2015).
681 49. LTD., T. U. Body Composition Analyzer BC-418MA Instruction Manual.
682 (2009).
683 50. Aulchenko, Y. S., Ripke, S., Isaacs, A. & van Duijn, C. M. GenABEL: an R
684 library for genome-wide association analysis. Bioinformatics 23, 1294–1296
685 (2007).
686 51. Willer, C. J., Li, Y. & Abecasis, G. R. METAL: Fast and efficient meta-
687 analysis of genomewide association scans. Bioinformatics 26, 2190–2191
688 (2010).
689 52. Ferreira, M. A. et al. Shared genetic origin of asthma, hay fever and eczema
690 elucidates allergic disease biology. Nat. Genet. 49, 1752–1757 (2017).
691 53. Benjamini, Y. & Hochberg, Y. Controlling the false discovery rate: a practical
692 and powerful approach to multiple testing. Journal of the Royal Statistical
693 Society 57, 289–300 (1995).
694 54. Magi, R., Lindgren, C. M. & Ã, A. P. M. Meta-Analysis of Sex-Specific
695 Genome-Wide Association Studies. 853, 846–853 (2010).
696 55. Wood, A. R. et al. Defining the role of common variation in the genomic and
697 biological architecture of adult human height. Nat. Genet. 46, 1173–1186
28 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
698 (2014).
699 56. Mägi, R. & Morris, A. P. GWAMA: Software for genome-wide association
700 meta-analysis. BMC Bioinformatics 11, (2010).
701
702 Figure legends
703 Figure 1. Segmental body impedance analyses use bioelectrical impedance to
704 estimate body composition: fat mass, muscle mass, etc. Adipose tissue mass, in this
705 study, had been estimated using the Tanita BC-418MA body composition analyzer
706 (a). This machine uses an eight-electrode method, which allows for five
707 measurements of impedance. Current is supplied to the front of both feet and the
708 fingertips of both hands. Voltage is measured on either heel or thenar portion of the
709 palms. Body composition is derived from a regression formula for each body part.
710 The formula is derived from regression analysis using height, weight, age and
711 impedance for each body part as predictors for composition of each body part as
712 assessed by DXA (b). GWAS for AFR, LFR and TFR were conducted in the UK
713 Biobank cohort and revealed associations with loci that have not previously been
714 associated with standard anthropometric traits. (c) shows a manhattan plot with
715 combined results for association studies of body fat ratios in combined and sex-
716 stratified analyses. Novel loci are highlighted in red.
717
718 Figure 2. Body fat ratio associated loci and the overlap of associations between
719 AFR, LFR and TFR. Loci are denoted by the nearest gene or by the most likely
720 target gene (see methods section).
721
722 Figure 3. Enrichment analyses of genes at LFR and TFR associated loci. (a)
29 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
723 Reconstituted gene sets found to be enriched for TFR- and LFR-associated genes (in
724 males and females) were compared with results from previous GWAS on
725 WHRadjBMI12, BMI9 and height55. Tissue and cell type enrichment of (b) TFR- and
726 (c) LFR-associated genes. Red bars denote tissues gene sets that were significantly
727 enriched for LFR- and TFR-associated genes at FDR < 0.05/12.
728
729 Supplementary Figure and Tables
730 Supplementary Table 1. Basic characteristics of UK Biobank participants 731 included in the analyses.
732 Supplementary Table 2. SNP-heritability (h2) and LD score regression intercept 733 estimates of body fat ratios.
734 Supplementary Table 3. Phenotypic correlation between body fat ratios and 735 anthropometric traits.
736 Supplementary Table 4. Genetic correlations between body fat ratios and 737 anthropometric traits.
738 Supplementary Table 5. Direction of effect for body fat ratio– 739 associated leading SNPs for standard anthropometric traits.
740 Supplementary Table 6. Overlap between body fat ratio- 741 associated loci with non-anthropometric trait-associations from 742 previous GWAS.
743 Supplementary Figure 1. Quantile-quantile (QQ) plots of the distribution of p- 744 values from GWAS of body fat ratios in the UK Bibank sex-combined, female 745 and male discovery cohorts.
746 Supplementary Figure 2. Manhattan plots for GWAS results for arm fat ratio in 747 the combined discovery cohort and sex-stratified analyses.
748 Supplementary Figure 3. Manhattan plots for GWAS results for leg fat ratio in 749 the combined discovery cohort and sex-stratified analyses.
750 Supplementary Figure 4. Manhattan plots for GWAS results for trunk fat ratio 751 in the combined discovery cohort and sex-stratified analyses.
752 Supplementary Figure 5. Overlap of enriched gene sets from 753 DEPICT for body fat ratios.
754
30 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
755 Supplementary data sets
756 Supplementary Data 1. Independent signals for association with body fat ratios.
757 Supplementary Data 2. Expression quantitative loci (eQTLS) in LD with leading
758 SNPs at body fat ratio-associated loci.
759 Supplementary Data Set 3. Results from enrichment analyses by DEPICT.
760
761
31 bioRxiv preprint
Tables
doi:
Table 1. Novel body fat ratio-associated loci. Entries represent loci that have not previously been associated with an anthropometric trait. certified bypeerreview)istheauthor/funder.Allrightsreserved.Noreuseallowedwithoutpermission. Discovery Replication Meta-analysis https://doi.org/10.1101/207498 Most likely Strongest Directio Locus Lead SNP MAF target gene associated trait β P-value β P-value P-value n - chr1:10,273,179 - α 7.03*10 rs4846204 12.7% KIF1B TFR - combined 0.035 2.87*10-8 0.026 2.25*10-9 ++ 10,370,712 16 - chr1:155,143,768 - α 3.62*10 rs4971091 37.7% KRTCAP2 AFR - combined -0.024 8.23*10-8 -0.013 1.35*10-5 -- 155,144,300* 11 - chr1:154,848,581 - α 7.52*10 rs180921974 2.3% PKLR TFR - combined -0.102 1.05*10-12 -0.097 9.45*10-24 -- 156,027,203 35 - α -8 -6 1.17*10 chr2:57,961,602 rs13011472 48.9% VRK2 AFR - combined -0.024 4.81*10 -0.013 6.96*10 11 -- ; this versionpostedJuly13,2018. - chr2:198,540,352 - α 8.07*10 rs148812496 49.7% RFTN AFR - combined -0.030 2.41*10-8 -0.015 4.19*10-5 -- 198,729,207 11 - chr3:48,939,080 - β 1.46*10 rs4521268 33.3% WDR6 TFR - combined -0.025 4.30*10-8 -0.021 5.00*10-12 -- 49,137,904 18 - chr3:50,034,637 - α 1.70*10 rs1986599 11.3% RBM6 TFR - combined -0.037 4.28*10-8 -0.021 1.47*10-6 -- 50,034,637 12 - β 1.13*10 chr3:52,718,183 rs565817433 43.3% NEK4 TFR - combined -0.023 6.83*10-8 -0.013 5.92*10-6 -- 11 - chr4:8,599,467 - α 3.97*10 rs2241069 46.1% CPZ TFR - females 0.033 1.56*10-8 0.019 7.86*10-7 ++ 8,604,153 13 -
chr5:74,689,249 - β 2.74*10 The copyrightholderforthispreprint(whichwasnot rs34341 42.3% ANKDD1B AFR - females -0.036 1.47*10-9 -0.025 1.20*10-10 -- 74,934,009 18 chr5:127,263,443 - α rs3812049 25.6% SLC12A2 AFR - males 0.061 9.24*10-18 0.021 1.99*10-6 2.40*10-6 ++ 127,562,880 - RP11- -8 -5 4.71*10 chr5:157,952,404 rs1317415 30.7% α LFR - females -0.034 6.11*10 -0.018 1.81*10 -- 32D16.1 11 - chr5:176,677,563 - α 1.67*10 rs34022431 2.6% NSD1 AFR - combined 0.071 8.55*10-8 0.040 7.12*10-6 ++ 176,732,485 11 - chr5:178,470,304 - α 7.50*10 rs888762 33.0% ADAMTS2 TFR - females 0.037 5.48*10-9 0.027 1.11*10-10 ++ 178,547,313 18 - chr6:34,539,910 - α 1.47*10 rs2492863 13.8% C6orf106 LFR - females -0.053 3.48*10-10 -0.046 5.26*10-16 -- 34,825,662 24 - chr7:20,374,338 - α 3.09*10 rs3823974 41.2% ITGB8 TFR - females -0.039 6.56*10-11 -0.028 2.43*10-12 -- 20,465,963 21 chr7:150,604,406 - rs56282717 24.3% KCNH2α AFR - combined -0.027 5.00*10-8 -0.020 5.59*10-9 3.03*10- --
32 bioRxiv preprint
150,657,209 15 - chr8:8,907,950 - α -8 -9 1.69*10 doi:
rs2044387 42.5% ERI1 AFR - females 0.035 1.14*10 0.024 1.03*10 ++ certified bypeerreview)istheauthor/funder.Allrightsreserved.Noreuseallowedwithoutpermission. 9,054,817 16 - https://doi.org/10.1101/207498 α 8.23*10 chr8:10,802,146 rs12546366 45.5% XKR6 AFR - females 0.031 9.29*10-8 0.021 6.50*10-8 ++ 14 - α 4.50*10 chr9:16,846,111 rs10962638 14.2% BNC2 TFR - combined -0.037 4.22*10-9 -0.014 3.90*10-4 -- 10 - chr9:86,582,923 - β 7.20*10 rs7039458 24.8% RMI1 , TFR - females 0.038 2.07*10-8 0.028 3.32*10-10 ++ 86,639,999 17 - chr9:129,672,095 - α 3.40*10 rs3780327 22.0% RALGPS1 LFR - combined -0.031 1.95*10-9 -0.016 2.50*10-6 -- 129,945,847 13 - chr10:72,412,776 - β 2.89*10 rs34821335 27.3% ADAMTS14 TFR - females 0.039 3.40*10-9 0.023 2.22*10-7 ++ 72,435,363 14 - ; chr10:96,009,182 - α 1.22*10 this versionpostedJuly13,2018. rs11289753 43.2% PLCE1 AFR - combined 0.030 2.41*10-12 0.021 1.60*10-12 ++ 96,039,597 22 - chr11:825,110 - α 6.65*10 rs1138714 43.2% AP006621.1 AFR - combined -0.024 5.47*10-8 -0.012 2.50*10-5 -- 825,777 11 chr11:65,715,204 - 8.29*10- rs71455793 4.6% TSGA10IP TFR - females -0.082 4.45*10-9 -0.039 1.85*10-5 -- 65,961,498 12 - chr11:69,449,076 - α 4.92*10 rs1789166 35.2% ORAOV1 AFR - combined -0.027 1.85*10-9 -0.013 2.05*10-5 -- 69,508,159 12 - chr12:50,468,643 - α 1.15*10 rs11614785 34.1% LARP4 TFR - females 0.040 1.37*10-10 0.026 2.93*10-10 ++ 51,220,897 18 - chr13:91,991,743 - α -8 -6 1.11*10 rs6492538 22.1% GPC5 TFR - females 0.039 3.16*10 0.021 8.33*10 11 ++
92,024,715 The copyrightholderforthispreprint(whichwasnot - chr14:50,923,249 - α 2.47*10 rs71420186 6.6% MAP4K5 LFR - combined -0.050 8.25*10-9 -0.038 3.20*10-11 -- 51,154,000 18 3.10*10- chr15:67,457,698 rs35874463 5.9% SMAD3 TFR - females 0.069 3.58*10-8 0.064 1.44*10-14 ++ 21 - chr15:89,349,708 - RP11- -9 -7 6.37*10 rs8026676 47.2% β TFR - females 0.034 7.24*10 0.020 2.71*10 ++ 89,363,866 343B18.2 14 chr15:100,692,953 4.91*10- rs72755233 11.3% ADAMTS17 TFR - females -0.074 5.09*10-16 -0.080 9.32*10-39 -- - 100,692,953 53 - chr16:31,075,175 - β 3.33*10 rs8050894 37.7% KAT8 AFR - combined -0.024 7.02*10-8 -0.017 4.76*10-9 -- 31,107,689 15 - chr16:69,547,741 - RP11- -9 -10 1.79*10 rs8057620 46.0% β AFR - females 0.035 2.27*10 0.024 4.40*10 ++ 69,887,720 419C5.2 17 - β 1.20*10 chr16:90,062,323 rs10584116 9.7% C16orf3 TFR - females -0.052 8.99*10-8 -0.033 6.97*10-7 -- 12
33 bioRxiv preprint
- chr17:42,278,916 - β 1.58*10 rs2071167 23.5% ASB16 LFR - combined -0.028 1.84*10-8 -0.023 1.05*10-6 -- 42,290,015 10 - doi: chr17:47,357,969 - α -10 -9 3.51*10 certified bypeerreview)istheauthor/funder.Allrightsreserved.Noreuseallowedwithoutpermission. rs28394864 46.2% ZNF652 TFR - females -0.036 8.89*10 -0.024 1.70*10 17 -- 47,454,515 https://doi.org/10.1101/207498 chr19:8,637,832 - 8.33*10- rs62621197 2.9% ADAMTS10 TFR - females -0.105 2.17*10-9 -0.147 3.45*10-39 -- 8,670,147 46 RP11- -8 -6 -6 chr20:6,408,920 rs6085551 47.7% α AFR - males 0.035 2.18*10 0.018 4.71*10 7.77*10 ++ 199O14.1 Leading SNP denotes the strongest associated SNP at each locus. The Most likely target gene -column denotes a gene related to the associated locus either by proximity to the leading SNP(α), LD (R2>0.8) with a leading eQTL SNP for the gene (β) or LD (R2>0.8) with a missense variant within the gene ( ). Independent signals within the same locus are denoted by *. β - effect size estimate per allele. Meta analyses were performed with METAL51. Direction - summary of effect direction for the discovery and replication cohorts, with one '+' or '-' per cohort.
;
this versionpostedJuly13,2018. The copyrightholderforthispreprint(whichwasnot
34 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
Table 2. Sex heterogeneous effects of body fat ratio-associated SNPs was assessed with GWAMA56 for all replicated SNPs (N=97). P-values denote the results from tests for heterogeneity between sexes. Bonferroni correction was used to correct for multiple testing and P-values < 0.05/97 (5.15*10-4) were considered significant. Effect Effect observed observed SNP Locus in females in males PAFR PLFR PTFR rs55750792 MFAP2γ + - - 3.14*10-4
rs2273368 WNT2B γ + - 1.59*10-7 1.82*10-8
rs11205303 SF3B4γ ++ + - 1.61*10-5 3.84*10-8
TSEN15, -8 -8 rs5779197 γ + - 8.76*10 2.48*10 GLT25D2 rs991967 TGFB2γ + - 1.83*10-4 5.61*10-5
rs2820443 LYPLAL1γ + - 3.00*10-6 2.11*10-5
rs10916174 ZNF678γ + - - 6.24*10-8
rs754537 RBJ, DNAJC27γ + 1.58*10-14 - -
rs3791679 EFEMP1γ + - 5.76*10-7 4.44*10-11
rs9853018 ZBTB38γ + - 1.31*10-16 2.29*10-26
rs4694504 ADAMTS3γ,β ++ + - - 4.14*10-4 rs994014 PRKG2γ + - 1.05*10-4 8.29*10-6
rs7680661 HHIPγ + - 5.36*10-10 8.59*10-14
rs34341 ANKDD1Bγ + 4.91*10-4 - -
rs115912456 VCANγ + - 4.58*10-5 2.23*10-4
rs3812049 SLC12A2α + ++ 2.92*10-7 4.65*10-4 - rs41271299 ID4γ + - 8.13*10-10 3.18*10-13
rs9358913 HIST1H2BEγ + - 1.23*10-13 1.29*10-12
rs9469762 HMGA1γ + - 3.67*10-6 -
rs2492863 C6orf106α + - 3.83*10-8 -
rs6570507 GPR126γ ++ + - - 6.10*10-5 rs798491 AMZ1γ + - 2.90*10-4 4.11*10-7
rs3823974 ITGB8α + - 1.84*10-3 1.73*10-4
rs481806 JAZF1γ + - - 2.48*10-6
rs527582137 CDK6γ + - - 1.60*10-4
rs35344761 PCSK5γ + - 3.07*10-8 5.70*10-9
rs11289753 PLCE1α + 6.94*10-5 - -
rs552846225 KDM2Aγ + - 3.66*10-6 1.71*10-8
rs11049566 CCDC91γ + - 3.06*10-3 5.13*10-4
rs11614785 LARP4α + - 6.27*10-5 4.01*10-6
rs12905253 PMLγ + - 6.56*10-4 1.96*10-4
rs11856122 ADAMTSL3γ + 0.52 1.08*10-18 2.46*10-19
rs28584580 ACANγ + - 2.96*10-11 7.31*10-15
rs3817428 ACANγ + - 5.99*10-10 3.02*10-9
rs72755233 ADAMTS17γ + - 4.19*10-8 8.87*10-11
rs4988781 ADAMTS17γ + - - 3.12*10-4
rs8050894 KAT8γ + 3.40*10-4 - -
rs55872725 FTOγ ++ + 6.96*10-6 - -
35 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
RBBP8, rs4800148 CABLES1, + - 3.76*10-10 1.14*10-10 γ C18orf45 rs62621197 ADAMTS10 + - - 3.54*10-11
rs2145270 BMP2γ + 2.87*10-4 - -
rs143384 GDF5γ + - 7.21*10-6 4.40*10-11
+ denotes an observed effect in one of the sexes, i.e. the SNP is associated (p < 1e-7) with one of the body fat ratios in that sex, ++ denotes that the effect was observed to be stronger in one sex when an effect was observed in both sexes. The effects were directionally consistent between sexes in all cases when SNP-effects could be observed in both sexes. The Locus column denotes a gene related to the associated locus either in annotation from previous GWAS(γ), LD (R2>0.8) with a leading eQTL SNP for the gene (β) or LD (R2>0.8) with a missense variant within the gene ( ).
36 bioRxiv preprint
Table 3. Body fat ratio-associated potentially damaging missense variants. LD w. doi:
Distance to leading certified bypeerreview)istheauthor/funder.Allrightsreserved.Noreuseallowedwithoutpermission.
Leading Position Missense Position leading SNP SNP Amino acid SIFT* Polyphen-2* https://doi.org/10.1101/207498 SNP Chr (bp) variant (bp) (bp) (R2) Gene substituion score score probably rs4733727 8 130,731,484 rs4144738 130,760,850 29,366 0.91 GSDMC Met/Thr deleterious damaging possibly rs180921974 1 155,268,131 rs72704117 155,175,089 93,042 0.95 THBS3 Ala/Thr deleterious damaging benign - possibly rs3817428 15 89,415,247 leading SNP ACAN Asp/Glu deleterious damaging rs72755233 15 100,692,953 leading SNP ADAMTS17 Thr/Ile deleterious benign
tolerated - possibly ; rs115912456 5 82,815,158 rs61749613 82,815,170 12 1.00 VCAN Lys/Glu this versionpostedJuly13,2018. deleterious damaging probably rs9358913 6 26,239,404 rs7766641 26,184,102 55,302 0.82 HIST1H2BE Gly/Ser tolerated damaging possibly rs351855 5 176,520,243 leading SNP FGFR4 Gly/Arg tolerated damaging possibly rs71420186 14 50,960,918 rs12881869 50,923,249 37,669 0.96 MAP4K5 Ala/Thr tolerated damaging benign - possibly rs5779197 1 184,009,826 rs2274432 184,020,945 11,119 0.96 TSEN15 Gly/Asp tolerated damaging GDF5; Ala/Ser; benign - possibly rs143384 20 34,025,756 rs224331 34,022,387 3,369 0.73 tolerated GDF5OS Ala/Glu damaging benign - possibly The copyrightholderforthispreprint(whichwasnot rs62621197 19 8,670,147 leading SNP ADAMTS10 Arg/Gln tolerated damaging *Polyphen-232 and SIFT31 scores were extracted from Ensembl (www.ensembl.org). SIFT utilizes protein sequence data to predict the effects of a missense variant while Polyphen-2 also considers protein structure and evolutionary conservation. Polyphen-2 generates a probability that a missense mutation is damaging, while SIFT generates a probability that the amino acid change is tolerated. Polyphen-2 uses a tree-tiered prediction score: 0-0.15 is predicted to be benign, 0.15 to 0.85 possibly damaging and 0.85 -1.0 more confidently predicted to be damaging (i.e., probably damaging). A SIFT score < 0.05 is predicted to be deleterious31. Polyphen-2 and SIFT scores are presented as ranges in the cases where they differ due to being part of different transcript variants.
37 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
a b
V Trunk Whole body
V V Right arm Left arm Right arm Left arm
Right Leg Left leg
V V c Right Leg Left leg
30
20
10
0 1 2 3 4 5 6 7 8 9 10 11 1213 14 16 18 20 22 bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
AFR
MC4R KCNH2 KAT8 ANKDD1B SOCS5 ZC3H4 AP006621.1 RP11-199014.1 XKR6 RFTN TMEM18 MAPK3 BCDIN3D/FAIM2 RP11-419C5.2 VRK2 MYEOV SEC16B TFAP2D DOCK3 NSD1 PLCE1 ORAOV1 ERI1 KRTCAP2 RBJ FTO CPZ ADAMTS10 SLC12A2 ADAMTSL3 ADAMTS2 BMP2 ASB16 ACAN HIST1H2BE C16orf3 FGFR4 C6orf106 ADAMTS14 HHIP CABLES1 CDC42ep3 ESR1 HMGA1 ADAMTS17 ITGB8 GDF5 CDK6 GPC5 RP11-32B16.1 KDM2A LIN28B SF3B4 TGFB2 JAZF1 FAM184B
KNTC1 ADAMTS3 TSEN15 WNT2B MLZE GPR126 RBM6 NEK4 SP6 MAP4K5 AMZ1 PML VCAN PRKG2 B4GALNT3 PKLR CCDC91 MAP3K1 KIF1B MTCH2 BNC2 PCSK5 GATAD2A DYM MFAP2 SH2B1 EFEMP1 ZBTB38 ID4 PDIA4 SMAD3 RALGPS1 RMI6 RP11-343B18.2 TSGA10IP WDR6 LFR ZNF652 ZNF678 TFR bioRxiv preprint doi: https://doi.org/10.1101/207498; this version posted July 13, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
a.
WHRadjBMI BMI LFR/TFR Height 112 801 3 11 26 74 0 0 125 2 6 20 7 39 73
b. c.
Myometrium Uterus Urogenital Female genetalia
Adipose tissue White adipose tissue Tissues Subcutaneous fat Abdominal fat Stomatognathic Sense organs
CNS
Joint Capsule Musculoskeletal Joints Synovial Membrane Cartilage Cervical vertebrae Hemic and Immune
Gonads Endocrine Endocrine glands
Digestive Chondrocytes Mesenchymal stem cells Fibroblasts Adipocytes Other Cells Osteoblasts
Arteries Cardiovascular
0 2 4 6 8 0 2 4 6 8 -Log10(P) -Log10(P)